Almost prime generators and almost integers

Here are two apparently unrelated things you may have seen before. The first is an observation going back to Euler that the polynomial

produces a long sequence of primes. Namely, the values are prime for n = 1, 2, 3, …, 40.

The second is that the number

is extraordinarily close to an integer. This number is known as Ramanujan’s constant. It differs from its nearest integer by 3 parts in 10³⁰. Ramanujan’s constant equals

262537412640768743.99999999999925…

There is a connection between these two facts: The polynomial

returns primes for n = 1, 2, 3, …, k − 1 primes if 4k − 1 is a Heegner number, and

is almost an integer if d is a (large) Heegner number.

Source: The Book of Numbers by Conway and Guy.

Heegner numbers

So what’s a Heegner number and how many are there? An integer d is a Heegner number if the ring generated by appending √−d to the integers has unique factorization. There are nine such numbers:

1, 2, 3, 7, 11, 19, 43, 67, 163.

There’s deeper stuff going on here than I understand—modular forms, the j-function, etc.—so this post won’t explain everything. There’s something unsatisfying about saying something is “almost” an integer without quantifying. There’s a way to be more precise, but we won’t go there. Instead, we’ll just play with the results.

Mathematica computation

First we look at the claim that n² − n + k produces primes for n = 1 through k − 1 if 4k − 1 is a Heegner number. The values of k such that 4k − 1 is a Heegner number are 2, 3, 5, 11, and 17. The following code shows that the claim is true for these values of k.

k = {2, 3, 5, 11, 17}
claim[x_] := AllTrue[
  Table[n^2 - n + x, {n, x - 1}], 
  PrimeQ
]
AllTrue[k, claim]

This returns True, so the claim is true.

As for exp(π √d) being close to an integer, this apparently only true for the last three Heegner numbers.

h = {1, 2, 3, 7, 11, 19, 43, 67, 163}
For[i = 1, i < 10, i++, 
  Print[
    AccountingForm[
      N[
        Exp[ Pi Sqrt[ h[[i]] ] ], 
        31
      ]
    ]
  ]
]

(The function AccountingForm suppresses scientific notation, making it easier to see where the decimal portion of the number starts.)

Here are the results:

                23.1406926327792
                85.0196952232072
               230.7645883191458
              4071.9320952252610
             33506.1430655924387
            885479.7776801543194
         884736743.9997774660349
      147197952743.9999986624548
262537412640768743.9999999999993

I manually edited the output to align the decimal points and truncate the decimal places beyond that needed to show that the last number is not an integer.